Precise air quality forecasting is a necessary requirement for protecting public health and in the design of environmental policies. Traditional systems for air quality prediction fail to capture the dynamic changes in the air because of its complexity. We propose here that a hybrid system advanced in machine learning techniques is used to resolve the issue. It presents some models for assessing: SVR, GA-KELM, DBN-BP, and introduces here Bi-LSTM to ensure effectiveness on the time-series data. The experimental results are then demonstrated to show a predominance of Bi-LSTM with smaller values of RMSE and MSE. The combination of GA-KELM makes the model's prediction capabilities more enhanced and robust for the Bi-LSTM model. The hybrid methodology here supports enhanced precision in the quality of air, thereby empowering improved decision-making in pollution control and protection of the health of the public.